fix(ignore-filter): tolerate ./-prefixed and empty relative paths in isIgnored

[tirth8205]

Problem

• isIgnored is typed as (relativePath: string): boolean — a total function that always returns a boolean. But it delegates straight to ignore@7's ig.ignores() without any normalization, and ig.ignores() THROWS on two classes of perfectly ordinary relative-path strings: (1) an empty string, and (2) a path with a leading './'. These are not exotic inputs: path.relative(projectRoot, projectRoot) returns "" (the root itself), and file-walkers / glob results very commonly emit './src/index.ts'-style relative paths. In both cases the caller does not get false (or true) back — the whole scan crashes with an uncaught exception. Verified at runtime against the package's pinned ignore@7.0.5: isIgnored("") throws path must not be empty; isIgnored("./src/a.ts") throws path should be a path.relative()'d string, but got "./src/a.ts". The documented contract (boolean return) is silently violated for…

Fix

• Normalize before delegating: treat falsy/empty as not-ignored and strip a leading './' so ignore@7 accepts it. e.g. isIgnored(relativePath: string): boolean { if (!relativePath) return false; const normalized = relativePath.replace(/^.//, ""); if (!normalized) return false; return ig.ignores(normalized); }, Verified this returns: isIgnored("") -> false, isIgnored("./dist/a.js") -> true, isIgnored("dist/a.js") -> true, isIgnored("src/a.ts") -> false. <30 LOC.

Testing

Adds unit test(s) that fail before the change and pass after. The full core test suite, eslint, and tsc --noEmit all pass locally on this branch.

Found via a static correctness audit of the ignore filter.

🤖 Generated with Claude Code

[thejesh23]

A few concerns before this lands.

1. The normalization doesn't cover the other shapes path.relative actually produces. ig.ignores() from ignore@7 also throws on "../foo" (path escapes root) and on Windows backslash paths like ".\\src\\a.ts" / "src\\a.ts" — path.relative(projectRoot, target) produces both. The PR fixes one stripping rule but the body claims the contract violation is fixed; on Windows or with out-of-root callers it still throws. Either reject these explicitly with false, convert separators / leading ../, or document the input constraint.

2. The regex is one-shot and order-sensitive. "./".replace(/^\.\//, "") yields "" (correctly caught by the second guard), but "././foo" becomes "./foo" — still in the ignore@7 reject set — and ".//foo" becomes "/foo", also rejected. Use a loop / a stronger normalization (path.posix.normalize after slash-conversion) so the function is closed over its own output.

3. Test gap on the negation. The new tests confirm ./dist/index.js is ignored and ./src/index.ts is not, but there's no test that the previously-throwing inputs from the PR body actually return rather than throw (expect(() => ...).not.toThrow()), and no coverage for "./" alone (the trim-to-empty path), which is the one branch a future refactor is most likely to drop.

Nit: the JSDoc on IgnoreFilter.isIgnored still says "given relative path" with no mention of the now-tolerated ./ prefix or empty-as-root semantics — worth one line so callers know what's contractually accepted.